27 July 2018, 18:54 | #1 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Furia Woes/Counterfeit Chips [Resolved tnx to Lotharek and Boris]
Hi all
[For the following, please substitute all instances of the word 'byte' with 'nibble' or 4 bits - my bad!] This is my first post here, so before i begin, hello to all. Nice community here, i've lurked for some time, lol. Anyway! I recently purchased a Furia from amigastore.eu and i've been having some problems with stability. I have no doubt that right now a dozen users in unison are shouting 'it's not seated properly!' or 'ground the damn thing!' - and while your suggestions are totally appropriate, I should say that i have tried *everything* I can find even the vaguest reference to online. Before i describe the problem in detail i'll just give a short list: So far I have:
I've looked at the 7mhz and the 33mhz clocks on the scope and they seem fine. I've a 100mhz scope so if anyone has any other ideas of things to check then let me know! The 5v/12v rails are both pretty flat - we're talking tens of mV if any visible ripple. Okay! So the problem is simply that the system crashes (usually to a guru) after anywhere from a couple of minutes to a whole hour of use. This happens in games, it happens in workbench. It happens when in heavy use (such as the Frontier intro) at best after about 20 minutes, at worst after about a minute. It happens with no use at all (left idle on the workbench desktop) after at best about an hour? The guru is usually an illegal operation, or illegal address, and the address is always in the Z2 space (which makes sense, as the fastram is always active). What is *really* interesting, is that when running MBRTest2, if the system doesn't crash beforehand, I do actually get test failures on the 8mb Z2 fastram range. This does not happen on the 1.5meg upper range used for maprom etc, nor does it happen with the CHIPRAM. When it *does* fail, it seems to fail on the MARCH/MSCAN test (though i'm still trying to get more data on this) - but what is really curious is that it *always* fails on the 1st, 3rd, or 6th byte of the 8byte test word. For example: a write of AAAAAAAA might return a 3A3AA3AA on read. Sometimes it's just the first byte thats incorrect, sometimes it's all three, but it is *never* any of the other five bytes. I've (as of today) started making a list of the address lines that MBRTest reports, and so far it is always four or more of: 8, 9, 10, 11, 20, 21, 22, 23, 28, 29, 30, 31 Most of the time it is about half of these, but occasionally it's all of them. I'm not *certain* of the applicability of this data to the Furia specifically, but i'm guessing this relates to only the 1st, 3rd, or 6th byte of the test words failing. Sometimes the scan runs repeatedly for around 5 or 10 minutes without an error - but it will always either crash first, or given an error in the 8meg fastram range only. I'm at my wits end with this! haha... i've emailed amigastore.eu about a week ago but so far received no reply. I emailed them again today in the hopes that they'll get back to me. Obviously i'm not certain that the furia is at fault, so i'm open to any and all suggestions! That being said, with the pattern of failures in MBR2 i'm starting to lean towards something being wrong with the furia? I'd appreciate *any* help whatsoever! Many thanks all! John Last edited by project23; 13 October 2018 at 18:30. Reason: Correcting byte/nibble |
27 July 2018, 19:35 | #2 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Ah crap - substitute byte with nibble, or 4 bits, or whatever
Ya know what I mean! Lol |
27 July 2018, 21:52 | #3 |
Registered User
Join Date: Sep 2007
Location: Stockholm
Posts: 4,332
|
Given your thorough testing, I'm prone to agree that the fault is on the Furia itself.
|
28 July 2018, 07:49 | #4 |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
But you have a good model, as it works for more than 30 minutes. My red 2.x board requires 30min of heatup with endless guru and reset cycles. After that it runs for another 30 minutes relative stable. Than it's to hot or something similar and give the same as it was to cold.
In the 30min functionality all demo disks, the frontier intro or playing pinball is no problem. Have nearby same machine with solid RT52B PSU and set to 5.1V against cable resistance. Have changed to SD-Card adapter but it also crashes from floppy bootup. Fiddled with maxtransfer, because i do not understand what the ide boost option in furiatune is on or off by default. If possible i would clock the beast down to get it more stable, the crashing experience is really annoying. |
28 July 2018, 15:00 | #5 | |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Quote:
If the Amiga itself is not to blame, then it is not a good board. When did you buy/receive yours? I got mine a few weeks ago, and it was purchased around the time Lotharek posted a video saying he had supply problems. I'm sure those boards were not sent out, but given that the problem is confined to what looks like specific address lines (if i'm understanding correctly) then has the supply problem truly been sorted? I dunno - What i do know is i'm still waiting for amigastore.eu to get back to me. Cheers, John By the way - i believe its something of a myth that this thing is overclocked. I initially thought so myself, since sysinfo gives 36mhz - but it also gives 37 something if i switch to NTSC mode, and something like 33 if i disable CPU caches. Using a scope to probe around the clock, the *only* clock signal running to the cpu I can see is 33.3mhz. The clock itself is 66mhz (you can check this yourself). Now without a schematic i can't be certain, but i'd be welling to bet that 33mhz goes straight to the CPU (and is perhaps mixed with the 7 to produce the 40mhz for the FPU?). This would be consistent with Lotharek's site (and amigastore), which states that the 020 is clocked at 33, and the FPU at 40. Neither site mention an overclock. The point being that the 020 and the FPU should be rock stable, especially with a heatsink on the 020 - unless the chip is an 020/25 and not an 020/33, but both sites are quite clear that it is a 68020/33. Last edited by project23; 28 July 2018 at 15:14. |
|
29 July 2018, 15:30 | #6 | |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
Quote:
The whole board is in my same album. As you see it's the /25 and so it is overclocked around 32% what also explains the amount of heat it produces. Again for all electronic experts, please help me to clock the Furia down to 25MHz, what must be done for that. I assume that helps for stability, or the CPU must be exchanged to a /33 type. For me it's around having fastram in my A600 and not ridiculous clocking. Solderbro |
|
29 July 2018, 18:04 | #7 | |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Quote:
That's a hell of an overclock. I know these chips clock well on an individual basis, but i sincerely doubt that they all clock reliably enough to sell en masse at such a high speed. If there was a software option or jumper setting to take the 25mhz up to 30 or 33 then i'd call that reasonable, but with the current board layout such a change is (unless i'm mistaken) impossible. Maybe it can be flashed to divide the 66mhz clock differently, if the division is done in the cpld, but i really wouldn't know. As far as i can see, the only way to clock this thing down would be to remove the 66mhz crystal and replace it with a 50mhz one. That'd take the FPU down to, i'm guessing, 33mhz, and the CPU to 25mhz. This is all a guess of course, as no schematics or documentation is available. Note: I have no plans to make changes to my own board as i assume it'll violate warranty. Cheers, John |
|
29 July 2018, 22:37 | #8 |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
Bought my card at Everdrive, they had also no information that it is overclocked on regular basis. Yes figured the 66.66 oscillator out and ordered a 50Mhz model, modern forms run at 3.3 to 5.0V. Think such resolder will result in a accelerator faster than the ACA0202 but same stability.
In stable version (hint for Lotharek) the Furia will be a fair budget accelerator, yes jumpers for the user would be nice. Don't know what the chip below the cooler left has to do, as the FPU runs without cooler on the other end. Will report after the rebuild. Cheers Solderbro |
29 July 2018, 23:09 | #9 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
(Quick reply as on my phone)
The chip on the left is a CPLD handling (i imagine?) the bus arbitration etc between the Amiga and the 020 etc... Its definitely a CPLD though, xilinx I think? |
30 July 2018, 03:23 | #10 |
Registered User
Join Date: Sep 2007
Location: Stockholm
Posts: 4,332
|
My white Furia has been very stable; I assume modern EC020s are quite tolerant to overclocking just like later masks of the 68040 and 68060. But it looks like there is a quality control problem here.
|
30 July 2018, 15:14 | #11 | |
Registered User
Join Date: Jul 2012
Location: Kassel
Posts: 160
|
Quote:
Thats a Fake Cpu Mask and Produktion Date are not possible. |
|
30 July 2018, 16:44 | #12 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Where can i access a list of cpu masks? Could you explain how you deduced that it's a fake one?
As for the production date, i think you might be right. According to wiki, at least, it was Rochester Electronics who re-established manufacturing of the chip in 2014. Per their online catalogue, they do not sell an MC68EC020/25 in a QFP, only a 16mhz variant. EDIT: Okay so I found a few news articles confirming that Freescale stopped manufacture of the 020 in 2010, then e2v took up support using a load of FS 020 wafers they purchased, but *this* was only available as ceramic QFP - not plastic. It was discontinued fully a few years ago. It seems after 2010 this was the *only* place you could get a new 020, until 2014 from Rochester Electronics, who unless i'm mistaken do not manufacture a 25mhz QFP of the EC020. I'm still curious about the CPU mask though? Could you shed some light on that please? Cheers! John BTW - Thanks for the heads up esel |
30 July 2018, 17:00 | #13 |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
Processor and FPU on my Furia have the Freescale Semiconductor logo. They build the licensed chips, today NXP as new owner offer the datasheet and is the latest producer: https://en.wikipedia.org/wiki/Freescale_Semiconductor
Could not find information about the stable clocking range Will see how stable it works with regular clocking. Solderbro Last edited by Solderbro; 30 July 2018 at 17:03. Reason: missing link |
30 July 2018, 17:07 | #14 | |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Quote:
Freescale stopped manufacture in 2010, NXPs website lists the 020 as no longer manufactured, and their datasheets are in the archive section. https://www.nxp.com/products/no-long...cessor:MC68020 I think esel is right, and this is a fake chip. In which case the overclocking tolerance of the FS/MC chips is irrelevant. I'm sure its a *decent* copy, as it does run code, but it probably just doesn't overclock as well. John EDIT: Also, i'm not sure how important it is, but you appear to have different ram chips than me, and those of the stock images online. I doubt thats a big issue though. Last edited by project23; 30 July 2018 at 17:25. |
|
30 July 2018, 17:39 | #15 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Solderbro: Turns out we have *exactly* the same CPU...
MC68EC020FG25 2F37E QQTN1509 No wonder we're having the same problems. John |
31 July 2018, 08:01 | #16 |
Develop HW
Join Date: Jul 2011
Location: Nitra / Slovakia
Posts: 101
|
Please send me a card for analysis, more info PM.
Boris |
01 August 2018, 03:13 | #17 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
Will do!
I'm a little concerned however about what that would do for my warranty with amigastore.eu - I feel like they should maybe 'okay' this first. The trouble is, they've still not replied to any of my emails. I'll try them again tomorrow. I can't call them - they're in spain! lol John |
01 August 2018, 22:03 | #18 |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
Operation done, following points changed.
-Replaced the oscillator from 66.66 to 50.00 MHz. Found that TXC BBz9A is also a fake mask, because BB types from TXC have more pins and complexity as the 4 pin model. -Did attach the heat sink on the 68EC020 with thermal compound and removed the adhesive tape Results: The CPU is now on 27.3 MHz and produces less heat than before. Stability is improved much, apart from 'old diesel' effects at cold start and few reboots. FPU Speed is unknown and not listed by sysinfo, tips welcome. * Diesel Effects: Minutes of reset loops, getting yellow or green screen at start or 80000005, 8000000B Guru. But if the card decided to start and run, it is stable like my ACA020 card. Would say this rework improved my card a lot |
02 August 2018, 02:13 | #19 |
Used Register
Join Date: Jul 2018
Location: Liverpool
Age: 41
Posts: 437
|
How long have you managed to run it without crashes?
Cheers, John |
04 August 2018, 10:18 | #20 |
Registered User
Join Date: Jun 2018
Location: Hamburg
Age: 55
Posts: 56
|
Did a full round of Desert Dream first, followed by one hour Frontier rolling intro. Played Pinball Dreams and Fantasies for 'socket earthquake'. 2 hour Disksalv scan for problems from earlier crashes did follow. Only the old diesel problems exist now, Furia is running a loop until it starts finally and does it's job stable. Especially we have 26°C in the rooms and the heat is okay for A600 passive convection. This was a 1.20€ improvement with the Euroquartz XO91 50.
It would be a improvement to give the user a jumper for the decision to oc or take normal speed. Cheers Solderbro |
Currently Active Users Viewing This Thread: 1 (0 members and 1 guests) | |
Thread Tools | |
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Amiga 600 Furia - DAT Memory performance and some tips for Furia users | Nibbler | Hardware pics | 10 | 10 April 2019 22:45 |
Furia (white pcb from Lotharek) latest fw/update? | emuola | support.Hardware | 9 | 30 March 2019 15:01 |
Arcade [Resolved] | Umpa | support.FS-UAE | 3 | 16 April 2016 00:12 |
slideshow with chicks by Boris Valeyo | s2325 | request.Demos | 33 | 25 February 2009 11:48 |
BTTR Disgrace! (resolved ;-) | icycool | request.Demos | 16 | 02 June 2002 02:33 |
|
|